A few days ago, several experts warned of the high energy cost of developing generative AI models. Now a new study reveals that these tools require massive amounts of water to function. OpenAI, developer of the renowned ChatGPT language generator, needed 1.3 million kWh to train its linguistic model, and a team of researchers from universities in Colorado and Texas has discovered water is also necessary to cool its data centers.
This new research reveals that training GPT-3, ChatGPT’s predecessor model, consumed 700,000 liters of water – the equivalent of what is required to cool a nuclear reactor. Moreover, the study authors reported that this water equivalent could produce 370 BMW cars, or 320 Tesla electric vehicles. For a simple conversation of 20-50 questions and answers, ChatGPT consumes 0.5 liters of water, which might not sound like a lot but adds up quickly with millions of users.
In March, tech icons such as Elon Musk and Steve Wozniak signed a manifesto requesting to pause developing AI for six months, citing potential risks not only for the environment but also to spread hate speech, disinformation, and even trigger nuclear war.
OpenAI is a research laboratory with a mission to ensure that artificial general intelligence (AGI) advances the public’s well-being. Founded in 2015, by technologists Sam Altman and Greg Brockman, OpenAI is backed by philanthropists, tech CEOs, and investors who seek to build safe AI and ensure it works for the benefit of humanity. OpenAI is situated in San Francisco and creates a wide range of projects including ChatGPT and GPT-3, algorithms designed to generate human-like text.
Elon Musk is the CEO of Tesla, SpaceX, and founder of the Boring Company. He is well known for his contribution to the field of artificial intelligence, advocating for the development of technology that benefits humans. He has become increasingly vocal about the potential risks and dangers posed by AI, which is why he endorsed the call for a pause on AI development for six months. Musk believes that a deadline should be set to make sure that AI systems can be developed responsibly and safely.